Skip to content

Validation types

Agent Karma recognizes the validations that actually catch AI mistakes:

TypeCatches
TestLogic bugs, broken behavior, regressions
BuildCompile errors, broken imports
LintStyle/quality issues, common foot-guns
Type CheckType errors, wrong shapes, bad API usage

(It also understands Security scans and an Other bucket.)

  • Automatic — when you run a validation in VS Code’s integrated terminal, shell integration detects the command type and result. The raw command string is classified, then discarded.
  • ManualAdd Validation Command (or the @agentkarma /verify chat command, or the ✅ Log a Validation button) records one when auto-detection can’t see it — e.g. AI run in a browser, or commands run in an external terminal.

Why “ran” matters more than “passed”

Section titled “Why “ran” matters more than “passed””

A check only earns points if it actually ran. A failed test still earns the “tests run” credit (you looked) — but you don’t get the “tests passed” bonus. You can’t score well by simply never testing. That’s the whole philosophy: validation discipline, honestly measured.